近年來台灣致力於發展高解析度光學遙測衛星，其衛星影像能提供防災資訊、地理觀測及國土規劃等多項應用，直至今日對於遙測領域有著不可或缺之重要性。目前台灣僅福衛五號仍於太空中執行遙測任務，在不久的將來，福衛八號即將被發射，為使福衛八號衛星影像正確呈現地表地物的形狀、位置和地理坐標，且提升相較於過往獲取衛星正射影像之效率，規劃自動化幾何校正流程。
本研究主要採用具有80個有理函數參數(Rational Polynomial Coefficients, RPCs)的有理函數模型，用於描述衛星影像幾何成像，其中衛星供應商提供原始的RPCs隱含系統偏差，需使用控制點對其參數進行優化。為使獲取控制點方法更有效率，規劃於相對於地理坐標系統之參考影像上選取相應影像控制點，其選用的參考影像空間解析度與待正射影像不同，為降低不同衛星影像間灰度值及幾何偏差，先對影像進行預處理消除兩影像間差異性，接著採用影像區塊匹配策略以獲取均勻分布之影像控制點，由此建構不同衛星感測器影像匹配流程，其流程中採用衛星影像微調特徵提取與描述模型結合特徵匹配模型作為自監督式深度學習演算法進行影像匹配，使演算法易識別衛星影像上具獨特性點特徵，並與傳統基於特徵匹配演算法比較應用於幾何糾正流程的穩定性。對於幾何糾正流程而言，根據影像控制點數量及分布情形，選擇影像坐標修正模型及計算模型參數，用以修正虛擬影像控制點坐標，為防止有理函數模型中80個有理函數係數過度擬合有理函數模型，以奇異值分解重新求解RPCs，建立新穎自動化衛星影像幾何校正流程，最後藉由人工控制點評估幾何校正成果，迭代幾何校正流程直至幾何校正成果收斂，並以包含不同地形地貌測試區評估自動化衛星影像幾何校正流程之適應性。
本研究流程針對不同地物地貌的衛星影像進行測試，所使用的待正射影像為2米空間解析度的福衛五號影像、參考影像為10米空間解析度的Sentinel-2影像。實驗結果顯示，自動化幾何校正流程不僅具穩定性且具適應性，所採用的基於自監督特徵演算法與其他傳統基於特徵演算法的幾何糾正成果相較具有優勢，幾何校正結果不論測試於何種試驗區，其精度評估成果在福衛五號2米空間解析度下誤差皆約為2-4像素。

Recently, Taiwan has devoted itself to the development of high-resolution optical remote sensing satellites. The high-resolution optical satellite images can provide many applications such as disaster information, geographic observation, and national spatial planning. Currently, only FORMOSAT-5 still executes the mission of remote sensing in outer space, and FORMOSAT-8 will be launched in the near future. FORMOSAT-8 is also a high-resolution satellite made by Taiwan. To correctly represent the image geometry on satellite images and improve the efficiency of satellite image ortho-rectification, a novel method is presented which is a fully-automatic satellite image orthorectification process.
A rational function model with 80 rational polynomial coefficients (RPCs) is utilized to describe the geometry of space imageries in this research. Generally, RPCs provided by the satellite vendors imply systematic biases and thus further optimization is required to reach surveying level accuracy. The ortho-rectification process is based on the use of ground control points (GCPs), whose quality has a high impact on the ortho-rectification results. Thus, obtaining accurate ground control points for optimizing RPCs is critical. Different from traditional labor-intensive methods, a novel image matching method is adopted to find image control points both on target images and an ortho-rectified reference image, which is the combination of feature detection and description model fine-tuned by satellite images, and feature matching model called self-supervised deep learning image matching algorithm. This strategy makes the ortho-rectification process become automatic, robust, and attempts to distinguish more salient features than traditional methods in satellite images. To ensure the stability of image ortho-rectification proceeded by image matching control points, the discrepancy between the orthoimage and the target images should be reduced by conducting image preprocessing. Then the image block matching strategy is used to obtain the uniformly distributed image control points. For the geometric correction process, the coordinate corrected model is selected according to the number of control points and distribution and the parameter of the model are calculated to correct the virtual image control point coordinate for the RPCs optimization. The singular value decomposition (SVD) is adopted for new RPCs recalculation to prevent the 80 RPCs from overfitting. Finally, the orthoimages are evaluated with the manual control point. The iterative automatic satellite image orthorectification process is stopped from the convergence of geometric correction results.
In the experiments, the adaptive of the automatic satellite image orthorectification process is evaluated using satellite images from testing areas with different features and landforms. The FORMOSAT-5 image as orthoimage with a spatial resolution of 2 meters, and the Sentinel-2 image as reference image with a spatial resolution of 10 meters are adopted. The results show that the automatic orthorectification process is stable and adaptable for all the cases in the experiments. In addition, the self-supervised deep learning image matching algorithm shows the outperformance for satellite image orthorectification compared with other traditional feature-based image matching ones. The quantity assessment is performed using root mean square error, and the accuracy of satellite image orthorectification result is 2 to 4 pixels under the 2-meter spatial resolution of FORMOSAT-5 images.

Bay, H., Ess, A., Tuytelaars, T., & Van Gool, L. (2008). Speeded-up robust features Computer vision and image understanding (CVIU), 110(3), 346-359.
Bentoutou, Y., Taleb, N., Kpalma, K., & Ronsin, J. (2005). An automatic image registration for applications in remote sensing. IEEE transactions on geoscience and remote sensing, 43(9),2127-2137.
Calonder, M., Lepetit, V., Strecha, C., & Fua, P. (2010). Brief: Binary robust independent elementary features. Proceedings of the European Conference on Computer Vision (ECCV).
Cao, J., & Fu, J. (2018). Estimation of rational polynomial coefficients based on singular value decomposition. Journal of Applied Remote Sensing, 12(4), 044003.
Chen, L.-C., Teo, T.-A., & Liu, C.-L. (2006). The geometrical comparisons of RSM and RFM for FORMOSAT-2 satellite images. Photogrammetric Engineering & Remote Sensing, 72(5), 573-579.
Chen, L., Rottensteiner, F., & Heipke, C. (2021). Feature detection and description for image matching: from hand-crafted design to deep learning. Geo-spatial Information Science, 24 (1), 58–74.
Chen, S., Liu, X., Zhang, H., & Yan, G. (2017). Target location method based on homography and scene matching for micro-satellite images. International Conference on Optical and Photonics Engineering (icOPEN 2016).
Christiansen, P. H., Kragh, M. F., Brodskiy, Y., & Karstoft, H. (2019). Unsuperpoint: End-to-end unsupervised interest point detector and descriptor. arXiv preprint arXiv:1907.04011.
Chui, H., & Rangarajan, A. (2003). A new point matching algorithm for non-rigid registration. Computer vision and image understanding, 89(2-3), 114-141.
Chum, O., & Matas, J. (2005). Matching with PROSAC-progressive sample consensus. 2005 IEEE computer society conference on computer vision and pattern recognition (CVPR'05),220-226.
Chum, O., Matas, J., & Kittler, J. (2003). Locally optimized RANSAC. Proceedings of the DAGM- Symposium, 236-243.
Csurka, G., Dance, C. R., & Humenberger, M. (2018). From handcrafted to deep local features. arXiv preprint arXiv:1807.10254.
DeTone, D., Malisiewicz, T., & Rabinovich, A. (2016). Deep image homography estimation. arXiv preprint arXiv:1606.03798.
DeTone, D., Malisiewicz, T., & Rabinovich, A. (2017). Toward geometric deep slam. arXiv preprint arXiv:1707.07410.
DeTone, D., Malisiewicz, T., & Rabinovich, A. (2018). Superpoint: Self-supervised interest point detection and description. Proceedings of the IEEE conference on computer vision and pattern recognition workshops.
Fischler, M. A., & Bolles, R. C. (1981). Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography. Communications of the ACM, 24(6), 381-395.
Fraser, C. S., & Hanley, H. B. (2003). Bias compensation in rational functions for IKONOS satellite imagery. Photogrammetric Engineering & Remote Sensing, 69(1), 53-57.
Govindu, V., & Shekhar, C. (1999). Alignment using distributions of local geometric properties. IEEE transactions on pattern analysis and machine intelligence, 21(10), 1031-1043.
Harris, C., & Stephens, M. (1988). A combined corner and edge detector. Alvey vision conference,147-151.
Hong, G., & Zhang, Y. (2007). Combination of feature-based and area-based image registration technique for high resolution remote sensing image. IEEE International Geoscience and Remote Sensing Symposium, 377-380.
Kim, T., & Im, Y.-J. (2003). Automatic satellite image registration by combination of matching and random sample consensus. IEEE transactions on geoscience and remote sensing, 41(5), 1111-1117.
Kim, T., Shin, D., & Lee, Y.-R. (2001). Development of a robust algorithm for transformation of a 3D object point onto a 2D image point for linear pushbroom imagery. Photogrammetric Engineering and Remote Sensing, 67(4), 449-452.
Kluger, F., Brachmann, E., Ackermann, H., Rother, C., Yang, M. Y., & Rosenhahn, B. (2020). Consac: Robust multi-model fitting by conditional sample consensus. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition.
Kratky, V. (1989). Rigorous photogrammetric processing of SPOT images at CCM Canada. ISPRS Journal of Photogrammetry and Remote Sensing, 44(2), 53-71.
Lin, T.-Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Piotr Dollár, Zitnick, C. L. (2014). Microsoft coco: Common objects in context. European conference on computer vision.
Lowe, D. G. (2004). Distinctive image features from scale-invariant keypoints. International journal of computer vision, 60(2), 91-110.
Ma, J., Zhao, J., & Yuille, A. L. (2015). Non-rigid point set registration by preserving global and local structures. IEEE Transactions on image Processing, 25(1), 53-64.
Moravec, H. P. (1980). Obstacle avoidance and navigation in the real world by a seeing robot rover.
Muja, M., & Lowe, D. G. (2009). Fast approximate nearest neighbors with automatic algorithm configuration. VISAPP.
Myronenko, A., & Song, X. (2010). Point set registration: Coherent point drift. IEEE transactions on pattern analysis and machine intelligence, 32(12), 2262-2275.
Nguyen, T., Chen, S. W., Shivakumar, S. S., Taylor, C. J., & Kumar, V. (2018). Unsupervised deep homography: A fast and robust homography estimation model. arXiv:1709.03966 (2017)
O'NEILL, M. (1988). The generation of epipolar synthetic stereo mates for SPOT images using a DEM. International Archives of Photogrammetry and Remote Sensing, 27, 587-598.
OGC. (2000). The Compendium of Controlled Extensions (CE) for the National Imagery Transmission Format (NITF). STDI-0002, Version2.1.
Pautrat, R., Larsson, V., Oswald, M. R., & Pollefeys, M. (2020). Online Invariance Selection for Local Feature Descriptors. European Conference on Computer Vision.
Pizer, S. M., Johnston, R. E., Ericksen, J. P., Yankaskas, B. C., & Muller, K. E. (1990). Contrast-limited adaptive histogram equalization: speed and effectiveness. Proceedings of the First Conference on Visualization in Biomedical Computing, 337-345.
Poli, D., & Toutin, T. (2012). Review of developments in geometric modelling for high resolution satellite pushbroom sensors. The Photogrammetric Record, 27(137), 58-73.
Poli, D., Zhang, L., & Gruen, A. (2004). Orientation of satellite and airborne imagery from multi-line pushbroom sensors with a rigorous sensor model. International Archives of Photogrammetry and Remote Sensing, 35(B1), 130-135.
Rao, Y. R., Prathapani, N., & Nagabhooshanam, E. (2014). Application of normalized cross correlation to image registration. International Journal of Research in Engineering and Technology, 12-16.
Rocco, I., Arandjelović, R., & Sivic, J. (2020). Efficient neighbourhood consensus networks via submanifold sparse convolutions. European Conference on Computer Vision.
Rocco, I., Cimpoi, M., Arandjelović, R., Torii, A., Pajdla, T., & Sivic, J. (2018). Neighbourhood consensus networks. Neural Information Processing Systems.
Rosten, E., & Drummond, T. (2006). Machine learning for high-speed corner detection. European conference on computer vision.
Rublee, E., Rabaud, V., Konolige, K., & Bradski, G. (2011). ORB: An efficient alternative to SIFT or SURF. IEEE International Conference on Computer Vision.
Sarlin, P.-E., DeTone, D., Malisiewicz, T., & Rabinovich, A. (2020). Superglue: Learning feature matching with graph neural networks. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 4938–4947.
Shi, J., & Tomasi, C. (1994).Good Features to Track. IEEE Conference on Computer Vision and Pattern Recognition.
Song, Y., Cai, L., Li, J., Tian, Y., & Li, M. (2020). SEKD: Self-evolving keypoint detection and description. arXiv preprint arXiv:2006.05077.
Tahoun, M., Hassanien, A. E., & Reulke, R. (2015). Registration of Optical and Radar Satellite Images Using Local Features and Non-rigid Geometric Transformations. Lecture Notes in Geoinformation and Cartography Series, 249–261.
Tang, J., Kim, H., Guizilini, V., Pillai, S., & Ambrus, R. (2019). Neural Outlier Rejection for Self-Supervised Keypoint Learning. International Conference on Learning Representations.
Tareen, S. A. K., & Saleem, Z. (2018). A comparative analysis of sift, surf, kaze, akaze, orb, and brisk. International conference on computing, mathematics and engineering technologies (iCoMET).
Ton, J., & Jain, A. K. (1989). Registering landsat images by point matching. IEEE transactions on geoscience and remote sensing, 27(5), 642-651.
Tong, X., Liu, S., & Weng, Q. (2010). Bias-corrected rational polynomial coefficients for high accuracy geo-positioning of QuickBird stereo imagery. ISPRS Journal of Photogrammetry and Remote Sensing, 65(2), 218-226.
Torr, P. H., & Zisserman, A. (2000). MLESAC: A new robust estimator with application to estimating image geometry. Computer vision and image understanding, 78(1), 138-156.
Toutin, T. (2003). Block bundle adjustment of Ikonos in-track images. International Journal of Remote Sensing, 24(4), 851-857.
Toutin, T. (2004). Geometric processing of remote sensing images: models, algorithms and methods. International Journal of Remote Sensing, 25(10), 1893-1924.
Ventura, A. D., Rampini, A., & Schettini, R. (1990). Image registration by recognition of corresponding structures. IEEE transactions on geoscience and remote sensing, 28(3), 305-314.
Wiesel, J. (1985). Digital image processing for orthophoto generation. Photogrammetria, 40(2), 69-76.
Wu, Z., & Xiao, X. (2011). Study on Histogram Equalization. International Symposium on Intelligence Information Processing and Trusted Computing.
Yang, N., Stumberg, L. v., Wang, R., & Cremers, D. (2020). D3vo: Deep depth, deep pose and deep uncertainty for monocular visual odometry. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 1281–1292.
Yang, Z., Dan, T., & Yang, Y. (2018). Multi-temporal remote sensing image registration using deep convolutional features. IEEE Access, 6, 38544-38555.
Yildirim, I., Demirtas, F., Gulmez, B., Leloglu, U., Yaman, M., & Guneyi, E. (2019). Comparison of image matching algorithms on satellite images taken in different seasons. Türkiye Ulusal Fotogrametri ve Uzaktan Algılama Birliği Teknik Sempozyumu , 25-27.
Zeng, R., Denman, S., Sridharan, S., & Fookes, C. (2018). Rethinking planar homography estimation using perspective fields. Asian Conference on Computer Vision.
Zhang, F., Yang, T., Liu, L., Liang, B., Bai, Y., & Li, J. (2020). Image-only real-time incremental uav image mosaic for multi-strip flight. IEEE Transactions on Multimedia, 23, 1410-1425.
Zheng, Y., & Doermann, D. (2006). Robust point matching for nonrigid shapes by preserving local neighborhood structures. IEEE transactions on pattern analysis and machine intelligence, 28(4), 643-649.
Zhou, Q., Sattler, T., & Leal-Taixe, L. (2021). Patch2Pix: Epipolar-Guided Pixel-Level Correspondences. arXiv preprint arXiv:2012.01909.
Zitova, B., & Flusser, J. (2003). Image registration methods: a survey. Image and Vision Computing, 21(11), 977-1000.
巫婉瑜. (2008). 弱交會幾何衛星影像之有理函數模型區域平差. 中央大學土木工程學系學位論文.
林俊良, 余憲政, 劉小菁, 張莉雪, 張立雨, 李彥玲, & 李品儀. (2019). 福衛五號影像與國土利用判識. 國土及公共治理季刊, 7(2), 60-69.
林義乾. (2006). 以影像控制區塊進行福衛二號衛星影像定位. 臺灣大學土木工程學研究所學位論文.
張智安, & 陳良健. (2003). EROS A 衛星影像幾何改正之研究. 航測及遙測學刊, 8(3), 73-94.
張智安, & 陳良健. (2007). 有理函數模式於高解析衛星影像幾何改正之應用. 航測及遙測學刊, 12(3), 257-272.
陳俊愷. (2011). 影像特徵點萃取與匹配應用於福衛二號影像幾何糾正. 國立臺灣師範大學地理學系學位論文.
陳信安. (2007). 衛星影像幾何校正控制點自動萃取與匹配之研究. 國立臺灣大學理學院地理環境資源學系學位論文.
廖揚清, & 蔡文龍. (2006). 福衛二號影像糾正及誤差探討. 航測及遙測學刊, 11(4), 427-438.